You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

681 lines
32 KiB

(*
* Copyright (c) Facebook, Inc. and its affiliates.
*
* This source code is licensed under the MIT license found in the
* LICENSE file in the root directory of this source tree.
*)
open! IStd
open IResult.Let_syntax
open PulseBasicInterface
open PulseDomainInterface
type arg_payload = AbstractValue.t * ValueHistory.t
type model =
caller_summary:Summary.t
-> callee_procname:Procname.t
-> Location.t
-> ret:Ident.t * Typ.t
-> AbductiveDomain.t
-> ExecutionDomain.t list PulseOperations.access_result
module Misc = struct
let shallow_copy model_desc dest_pointer_hist src_pointer_hist : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model model_desc; location; in_call= []} in
let* astate, obj = PulseOperations.eval_access location src_pointer_hist Dereference astate in
let* astate, obj_copy = PulseOperations.shallow_copy location obj astate in
let+ astate =
PulseOperations.write_deref location ~ref:dest_pointer_hist
~obj:(fst obj_copy, event :: snd obj_copy)
astate
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.havoc_id ret_id [event] astate in
[ExecutionDomain.ContinueProgram astate]
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let early_exit : model =
fun ~caller_summary:_ ~callee_procname:_ _ ~ret:_ astate ->
Ok [ExecutionDomain.ExitProgram astate]
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let return_int : Int64.t -> model =
fun i64 ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let i = IntLit.of_int64 i64 in
let ret_addr = AbstractValue.Constants.get_int i in
let astate = PulseArithmetic.and_eq_int (Immediate {location; history= []}) ret_addr i astate in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.write_id ret_id (ret_addr, []) astate |> PulseOperations.ok_continue
let return_unknown_size : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let ret_addr = AbstractValue.mk_fresh () in
let astate =
PulseArithmetic.and_nonnegative (Immediate {location; history= []}) ret_addr astate
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.write_id ret_id (ret_addr, []) astate |> PulseOperations.ok_continue
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let skip : model =
fun ~caller_summary:_ ~callee_procname:_ _ ~ret:_ astate -> PulseOperations.ok_continue astate
let nondet ~fn_name : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model fn_name; location; in_call= []} in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.havoc_id ret_id [event] astate |> PulseOperations.ok_continue
let id_first_arg arg_access_hist : model =
fun ~caller_summary:_ ~callee_procname:_ _ ~ret astate ->
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.write_id (fst ret) arg_access_hist astate |> PulseOperations.ok_continue
end
module C = struct
let free deleted_access : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
(* NOTE: we could introduce a case-split explicitly on =0 vs ≠0 but instead only act on what we
currently know about the value. This is purely to avoid contributing to path explosion. *)
(* freeing 0 is a no-op *)
if PulseArithmetic.is_known_zero astate (fst deleted_access) then
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.ok_continue astate
else
let+ astate = PulseOperations.invalidate location Invalidation.CFree deleted_access astate in
[ExecutionDomain.ContinueProgram astate]
let malloc _ : model =
fun ~caller_summary:_ ~callee_procname location ~ret:(ret_id, _) astate ->
let hist =
[ValueHistory.Allocation {f= Model (Procname.to_string callee_procname); location}]
in
let immediate_hist = Trace.Immediate {location; history= hist} in
let ret_addr = AbstractValue.mk_fresh () in
let ret_value = (ret_addr, hist) in
let astate = PulseOperations.write_id ret_id ret_value astate in
let astate_alloc =
PulseArithmetic.and_positive immediate_hist ret_addr astate
|> PulseOperations.allocate callee_procname location ret_value
in
let+ astate_null =
PulseArithmetic.and_eq_int immediate_hist ret_addr IntLit.zero astate
|> PulseOperations.invalidate location (ConstantDereference IntLit.zero) ret_value
in
[ExecutionDomain.ContinueProgram astate_alloc; ExecutionDomain.ContinueProgram astate_null]
let malloc_not_null _ : model =
fun ~caller_summary:_ ~callee_procname location ~ret:(ret_id, _) astate ->
let ret_addr = AbstractValue.mk_fresh () in
let hist =
[ValueHistory.Allocation {f= Model (Procname.to_string callee_procname); location}]
in
let ret_value = (ret_addr, hist) in
let immediate_hist = Trace.Immediate {location; history= hist} in
let astate = PulseOperations.write_id ret_id ret_value astate in
PulseOperations.allocate callee_procname location ret_value astate
|> PulseArithmetic.and_positive immediate_hist ret_addr
|> PulseOperations.ok_continue
let cf_bridging_release access : model =
fun ~caller_summary:_ ~callee_procname:_ _ ~ret:(ret_id, _) astate ->
let astate = PulseOperations.write_id ret_id access astate in
PulseOperations.remove_allocation_attr (fst access) astate |> PulseOperations.ok_continue
end
module Cplusplus = struct
let delete deleted_access : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let+ astate =
PulseOperations.invalidate location Invalidation.CppDelete deleted_access astate
in
[ExecutionDomain.ContinueProgram astate]
let placement_new actuals : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "<placement new>()"; location; in_call= []} in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
( match List.rev actuals with
| ProcnameDispatcher.Call.FuncArg.{arg_payload= address, hist} :: _ ->
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.write_id ret_id (address, event :: hist) astate
| _ ->
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.havoc_id ret_id [event] astate )
|> PulseOperations.ok_continue
end
module StdAtomicInteger = struct
let internal_int =
Fieldname.make
(Typ.CStruct (QualifiedCppName.of_list ["std"; "atomic"]))
"__infer_model_backing_int"
let load_backing_int location this astate =
let* astate, obj = PulseOperations.eval_access location this Dereference astate in
let* astate, int_addr =
PulseOperations.eval_access location obj (FieldAccess internal_int) astate
in
let+ astate, int_val = PulseOperations.eval_access location int_addr Dereference astate in
(astate, int_addr, int_val)
let constructor this_address init_value : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let event = ValueHistory.Call {f= Model "std::atomic::atomic()"; location; in_call= []} in
let this = (AbstractValue.mk_fresh (), [event]) in
let* astate, int_field =
PulseOperations.eval_access location this (FieldAccess internal_int) astate
in
let* astate = PulseOperations.write_deref location ~ref:int_field ~obj:init_value astate in
let+ astate = PulseOperations.write_deref location ~ref:this_address ~obj:this astate in
[ExecutionDomain.ContinueProgram astate]
let arith_bop prepost location event ret_id bop this operand astate =
let* astate, int_addr, (old_int, old_int_hist) = load_backing_int location this astate in
let astate, (new_int, hist) =
PulseArithmetic.eval_binop location bop (AbstractValueOperand old_int) operand old_int_hist
astate
in
let+ astate =
PulseOperations.write_deref location ~ref:int_addr ~obj:(new_int, event :: hist) astate
in
let ret_int = match prepost with `Pre -> new_int | `Post -> old_int in
PulseOperations.write_id ret_id (ret_int, event :: hist) astate
let fetch_add this (increment, _) _memory_ordering : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::atomic::fetch_add()"; location; in_call= []} in
let+ astate =
arith_bop `Post location event ret_id (PlusA None) this (AbstractValueOperand increment)
astate
in
[ExecutionDomain.ContinueProgram astate]
let fetch_sub this (increment, _) _memory_ordering : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::atomic::fetch_sub()"; location; in_call= []} in
let+ astate =
arith_bop `Post location event ret_id (MinusA None) this (AbstractValueOperand increment)
astate
in
[ExecutionDomain.ContinueProgram astate]
let operator_plus_plus_pre this : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::atomic::operator++()"; location; in_call= []} in
let+ astate =
arith_bop `Pre location event ret_id (PlusA None) this (LiteralOperand IntLit.one) astate
in
[ExecutionDomain.ContinueProgram astate]
let operator_plus_plus_post this _int : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event =
ValueHistory.Call {f= Model "std::atomic<T>::operator++(T)"; location; in_call= []}
in
let+ astate =
arith_bop `Post location event ret_id (PlusA None) this (LiteralOperand IntLit.one) astate
in
[ExecutionDomain.ContinueProgram astate]
let operator_minus_minus_pre this : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::atomic::operator--()"; location; in_call= []} in
let+ astate =
arith_bop `Pre location event ret_id (MinusA None) this (LiteralOperand IntLit.one) astate
in
[ExecutionDomain.ContinueProgram astate]
let operator_minus_minus_post this _int : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event =
ValueHistory.Call {f= Model "std::atomic<T>::operator--(T)"; location; in_call= []}
in
let+ astate =
arith_bop `Post location event ret_id (MinusA None) this (LiteralOperand IntLit.one) astate
in
[ExecutionDomain.ContinueProgram astate]
let load_instr model_desc this _memory_ordering_opt : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model model_desc; location; in_call= []} in
let+ astate, _int_addr, (int, hist) = load_backing_int location this astate in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id ret_id (int, event :: hist) astate in
[ExecutionDomain.ContinueProgram astate]
let load = load_instr "std::atomic<T>::load()"
let operator_t = load_instr "std::atomic<T>::operator_T()"
let store_backing_int location this_address new_value astate =
let* astate, this = PulseOperations.eval_access location this_address Dereference astate in
let* astate, int_field =
PulseOperations.eval_access location this (FieldAccess internal_int) astate
in
PulseOperations.write_deref location ~ref:int_field ~obj:new_value astate
let store this_address (new_value, new_hist) _memory_ordering : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let event = ValueHistory.Call {f= Model "std::atomic::store()"; location; in_call= []} in
let+ astate = store_backing_int location this_address (new_value, event :: new_hist) astate in
[ExecutionDomain.ContinueProgram astate]
let exchange this_address (new_value, new_hist) _memory_ordering : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::atomic::exchange()"; location; in_call= []} in
let* astate, _int_addr, (old_int, old_hist) = load_backing_int location this_address astate in
let+ astate = store_backing_int location this_address (new_value, event :: new_hist) astate in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id ret_id (old_int, event :: old_hist) astate in
[ExecutionDomain.ContinueProgram astate]
end
module JavaObject = struct
(* naively modeled as shallow copy. *)
let clone src_pointer_hist : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "Object.clone"; location; in_call= []} in
let* astate, obj = PulseOperations.eval_access location src_pointer_hist Dereference astate in
let+ astate, obj_copy = PulseOperations.shallow_copy location obj astate in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id ret_id (fst obj_copy, event :: snd obj_copy) astate in
[ExecutionDomain.ContinueProgram astate]
end
module StdBasicString = struct
let internal_string =
Fieldname.make
(Typ.CStruct (QualifiedCppName.of_list ["std"; "basic_string"]))
"__infer_model_backing_string"
let internal_string_access = HilExp.Access.FieldAccess internal_string
let to_internal_string location bstring astate =
PulseOperations.eval_access location bstring internal_string_access astate
let data this_hist : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:(ret_id, _) astate ->
let event = ValueHistory.Call {f= Model "std::basic_string::data()"; location; in_call= []} in
let* astate, string_addr_hist = to_internal_string location this_hist astate in
let+ astate, (string, hist) =
PulseOperations.eval_access location string_addr_hist Dereference astate
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id ret_id (string, event :: hist) astate in
[ExecutionDomain.ContinueProgram astate]
let destructor this_hist : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let model = CallEvent.Model "std::basic_string::~basic_string()" in
let call_event = ValueHistory.Call {f= model; location; in_call= []} in
let* astate, (string_addr, string_hist) = to_internal_string location this_hist astate in
let string_addr_hist = (string_addr, call_event :: string_hist) in
let* astate = PulseOperations.invalidate_deref location CppDelete string_addr_hist astate in
let+ astate = PulseOperations.invalidate location CppDelete string_addr_hist astate in
[ExecutionDomain.ContinueProgram astate]
end
module StdFunction = struct
let operator_call lambda_ptr_hist actuals : model =
fun ~caller_summary ~callee_procname:_ location ~ret astate ->
let havoc_ret (ret_id, _) astate =
let event = ValueHistory.Call {f= Model "std::function::operator()"; location; in_call= []} in
[PulseOperations.havoc_id ret_id [event] astate]
in
let* astate, (lambda, _) =
PulseOperations.eval_access location lambda_ptr_hist Dereference astate
in
let* astate = PulseOperations.Closures.check_captured_addresses location lambda astate in
match AddressAttributes.get_closure_proc_name lambda astate with
| None ->
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
(* we don't know what proc name this lambda resolves to *)
Ok (havoc_ret ret astate |> List.map ~f:ExecutionDomain.continue)
| Some callee_proc_name ->
let actuals =
List.map actuals ~f:(fun ProcnameDispatcher.Call.FuncArg.{arg_payload; typ} ->
(arg_payload, typ) )
in
PulseOperations.call ~caller_summary location callee_proc_name ~ret ~actuals
~formals_opt:None astate
end
module GenericArrayBackedCollection = struct
let field =
Fieldname.make
(Typ.CStruct (QualifiedCppName.of_list ["__infer_pulse_model"]))
"__infer_model_backing_array"
let access = HilExp.Access.FieldAccess field
let eval location collection astate =
PulseOperations.eval_access location collection access astate
let eval_element location internal_array index astate =
PulseOperations.eval_access location internal_array (ArrayAccess (Typ.void, index)) astate
let element location collection index astate =
let* astate, internal_array = eval location collection astate in
eval_element location internal_array index astate
end
module GenericArrayBackedCollectionIterator = struct
let internal_pointer =
Fieldname.make
(Typ.CStruct (QualifiedCppName.of_list ["__infer_pulse_model"]))
"__infer_model_backing_pointer"
let internal_pointer_access = HilExp.Access.FieldAccess internal_pointer
let to_internal_pointer location iterator astate =
PulseOperations.eval_access location iterator internal_pointer_access astate
let constructor ~desc this init : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let event = ValueHistory.Call {f= Model desc; location; in_call= []} in
let* astate, (arr_addr, arr_hist) = GenericArrayBackedCollection.eval location init astate in
let* astate =
PulseOperations.write_field location ~ref:this GenericArrayBackedCollection.field
~obj:(arr_addr, event :: arr_hist)
astate
in
let* astate, (p_addr, p_hist) = to_internal_pointer location init astate in
PulseOperations.write_field location ~ref:this internal_pointer
~obj:(p_addr, event :: p_hist)
astate
>>| ExecutionDomain.continue >>| List.return
let operator_star ~desc iter ~caller_summary:_ ~callee_procname:_ location ~ret astate =
let event = ValueHistory.Call {f= Model desc; location; in_call= []} in
let* astate, iter_index = to_internal_pointer location iter astate in
let+ astate, (elem_val, elem_hist) =
GenericArrayBackedCollection.element location iter (fst iter_index) astate
in
let astate = PulseOperations.write_id (fst ret) (elem_val, event :: elem_hist) astate in
[ExecutionDomain.ContinueProgram astate]
let operator_plus_plus ~desc iter ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate =
let event = ValueHistory.Call {f= Model desc; location; in_call= []} in
let index_next = AbstractValue.mk_fresh () in
let* astate, current_index = to_internal_pointer location iter astate in
let* astate, element =
GenericArrayBackedCollection.element location iter (fst current_index) astate
in
(* Iterator is invalid if the value it points to is invalid *)
let* astate, _ = PulseOperations.eval_access location element Dereference astate in
PulseOperations.write_field location ~ref:iter internal_pointer
~obj:(index_next, [event])
astate
>>| ExecutionDomain.continue >>| List.return
end
module StdVector = struct
let reallocate_internal_array trace vector vector_f location astate =
let* astate, array_address = GenericArrayBackedCollection.eval location vector astate in
PulseOperations.invalidate_array_elements location (StdVector vector_f) array_address astate
>>= PulseOperations.invalidate_deref location (StdVector vector_f) array_address
>>= PulseOperations.havoc_field location vector GenericArrayBackedCollection.field trace
let init_list_constructor this init_list ~caller_summary:_ ~callee_procname:_ location ~ret:_
astate =
let event = ValueHistory.Call {f= Model "std::vector::vector()"; location; in_call= []} in
let* astate, init_copy = PulseOperations.shallow_copy location init_list astate in
let+ astate =
PulseOperations.write_field location ~ref:this GenericArrayBackedCollection.field
~obj:(fst init_copy, event :: snd init_copy)
astate
in
[ExecutionDomain.ContinueProgram astate]
let invalidate_references vector_f vector : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let crumb =
ValueHistory.Call
{ f= Model (Format.asprintf "%a()" Invalidation.pp_std_vector_function vector_f)
; location
; in_call= [] }
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
reallocate_internal_array [crumb] vector vector_f location astate
>>| ExecutionDomain.continue >>| List.return
let at ~desc vector index : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret astate ->
let event = ValueHistory.Call {f= Model desc; location; in_call= []} in
let+ astate, (addr, hist) =
GenericArrayBackedCollection.element location vector (fst index) astate
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id (fst ret) (addr, event :: hist) astate in
[ExecutionDomain.ContinueProgram astate]
let vector_begin vector iter : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let event = ValueHistory.Call {f= Model "std::vector::begin()"; location; in_call= []} in
let index_zero = AbstractValue.mk_fresh () in
let astate =
PulseArithmetic.and_eq_int (Immediate {location; history= []}) index_zero IntLit.zero astate
in
let* astate, ((arr_addr, hist) as arr) =
GenericArrayBackedCollection.eval location vector astate
in
let* astate, _ = GenericArrayBackedCollection.eval_element location arr index_zero astate in
PulseOperations.write_field location ~ref:iter GenericArrayBackedCollection.field
~obj:(arr_addr, event :: hist)
astate
>>= PulseOperations.write_field location ~ref:iter
GenericArrayBackedCollectionIterator.internal_pointer
~obj:(index_zero, [event])
>>| ExecutionDomain.continue >>| List.return
let reserve vector : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let crumb = ValueHistory.Call {f= Model "std::vector::reserve()"; location; in_call= []} in
reallocate_internal_array [crumb] vector Reserve location astate
>>| AddressAttributes.std_vector_reserve (fst vector)
>>| ExecutionDomain.continue >>| List.return
let push_back vector : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret:_ astate ->
let crumb = ValueHistory.Call {f= Model "std::vector::push_back()"; location; in_call= []} in
if AddressAttributes.is_std_vector_reserved (fst vector) astate then
(* assume that any call to [push_back] is ok after one called [reserve] on the same vector
(a perfect analysis would also make sure we don't exceed the reserved size) *)
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
PulseOperations.ok_continue astate
else
(* simulate a re-allocation of the underlying array every time an element is added *)
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
reallocate_internal_array [crumb] vector PushBack location astate
>>| ExecutionDomain.continue >>| List.return
end
module JavaCollection = struct
let set coll index new_elem : model =
fun ~caller_summary:_ ~callee_procname:_ location ~ret astate ->
let event = ValueHistory.Call {f= Model "Collection.set"; location; in_call= []} in
let* astate, ((old_addr, old_hist) as old_elem) =
GenericArrayBackedCollection.element location coll (fst index) astate
in
let+ astate =
PulseOperations.write_deref location ~ref:new_elem
~obj:(old_addr, ValueHistory.Assignment location :: old_hist)
astate
>>= PulseOperations.invalidate_deref location (StdVector Assign) old_elem
in
[pulse] Distinguish exit state at top level Summary: This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`): ``` type exec_state = | ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *) | ExitProgram of PulseAbductiveDomain.t (** represents the state originating at exit/divergence. *) ``` Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state. The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit. TODO: - Impurity analysis needs to be improved to consider functions that simply exit as impure. - The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?). Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely. Reviewed By: jvillard Differential Revision: D20791747 fbshipit-source-id: df9e5445a
5 years ago
let astate = PulseOperations.write_id (fst ret) (old_addr, event :: old_hist) astate in
[ExecutionDomain.ContinueProgram astate]
end
module StringSet = Caml.Set.Make (String)
module ProcNameDispatcher = struct
let dispatch : (Tenv.t, model, arg_payload) ProcnameDispatcher.Call.dispatcher =
let open ProcnameDispatcher.Call in
let match_builtin builtin _ s = String.equal s (Procname.get_method builtin) in
let pushback_modeled =
StringSet.of_list
["add"; "addAll"; "append"; "delete"; "remove"; "replace"; "poll"; "put"; "putAll"]
in
make_dispatcher
[ +match_builtin BuiltinDecl.free <>$ capt_arg_payload $--> C.free
; +match_builtin BuiltinDecl.malloc <>$ capt_arg_payload $--> C.malloc
; +match_builtin BuiltinDecl.__delete <>$ capt_arg_payload $--> Cplusplus.delete
; +match_builtin BuiltinDecl.__placement_new &++> Cplusplus.placement_new
; +match_builtin BuiltinDecl.objc_cpp_throw <>--> Misc.early_exit
; +match_builtin BuiltinDecl.__cast <>$ capt_arg_payload $+...$--> Misc.id_first_arg
; +match_builtin BuiltinDecl.abort <>--> Misc.early_exit
; +match_builtin BuiltinDecl.exit <>--> Misc.early_exit
; +match_builtin BuiltinDecl.__infer_initializer_list
<>$ capt_arg_payload $+...$--> Misc.id_first_arg
; +PatternMatch.implements_lang "System" &:: "exit" <>--> Misc.early_exit
; +match_builtin BuiltinDecl.__get_array_length <>--> Misc.return_unknown_size
; (* consider that all fbstrings are small strings to avoid false positives due to manual
ref-counting *)
-"folly" &:: "fbstring_core" &:: "category" &--> Misc.return_int Int64.zero
; -"folly" &:: "DelayedDestruction" &:: "destroy" &--> Misc.skip
; -"folly" &:: "Optional" &:: "reset" &--> Misc.skip
; -"folly" &:: "SocketAddress" &:: "~SocketAddress" &--> Misc.skip
; -"std" &:: "basic_string" &:: "data" <>$ capt_arg_payload $--> StdBasicString.data
; -"std" &:: "basic_string" &:: "~basic_string" <>$ capt_arg_payload
$--> StdBasicString.destructor
; -"std" &:: "function" &:: "operator()" $ capt_arg_payload $++$--> StdFunction.operator_call
; -"std" &:: "function" &:: "operator=" $ capt_arg_payload $+ capt_arg_payload
$--> Misc.shallow_copy "std::function::operator="
; +PatternMatch.implements_lang "Object" &:: "clone" $ capt_arg_payload $--> JavaObject.clone
; ( +PatternMatch.implements_lang "System"
&:: "arraycopy" $ capt_arg_payload $+ any_arg $+ capt_arg_payload
$+...$--> fun src dest -> Misc.shallow_copy "System.arraycopy" dest src )
; -"std" &:: "atomic" &:: "atomic" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdAtomicInteger.constructor
; -"std" &:: "__atomic_base" &:: "fetch_add" <>$ capt_arg_payload $+ capt_arg_payload
$+ capt_arg_payload $--> StdAtomicInteger.fetch_add
; -"std" &:: "__atomic_base" &:: "fetch_sub" <>$ capt_arg_payload $+ capt_arg_payload
$+ capt_arg_payload $--> StdAtomicInteger.fetch_sub
; -"std" &:: "__atomic_base" &:: "exchange" <>$ capt_arg_payload $+ capt_arg_payload
$+ capt_arg_payload $--> StdAtomicInteger.exchange
; -"std" &:: "__atomic_base" &:: "load" <>$ capt_arg_payload $+? capt_arg_payload
$--> StdAtomicInteger.load
; -"std" &:: "__atomic_base" &:: "store" <>$ capt_arg_payload $+ capt_arg_payload
$+ capt_arg_payload $--> StdAtomicInteger.store
; -"std" &:: "__atomic_base" &:: "operator++" <>$ capt_arg_payload
$--> StdAtomicInteger.operator_plus_plus_pre
; -"std" &:: "__atomic_base" &:: "operator++" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdAtomicInteger.operator_plus_plus_post
; -"std" &:: "__atomic_base" &:: "operator--" <>$ capt_arg_payload
$--> StdAtomicInteger.operator_minus_minus_pre
; -"std" &:: "__atomic_base" &:: "operator--" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdAtomicInteger.operator_minus_minus_post
; -"std" &:: "__atomic_base"
&::+ (fun _ name -> String.is_prefix ~prefix:"operator_" name)
<>$ capt_arg_payload $+? capt_arg_payload $--> StdAtomicInteger.operator_t
; -"std" &:: "integral_constant" < any_typ &+ capt_int
>::+ (fun _ name -> String.is_prefix ~prefix:"operator_" name)
<>--> Misc.return_int
; -"std" &:: "vector" &:: "vector" <>$ capt_arg_payload
$+ capt_arg_payload_of_typ (-"std" &:: "initializer_list")
$+...$--> StdVector.init_list_constructor
; -"std" &:: "__wrap_iter" &:: "__wrap_iter" <>$ capt_arg_payload $+ capt_arg_payload
$+...$--> GenericArrayBackedCollectionIterator.constructor ~desc:"iterator constructor"
; -"std" &:: "__wrap_iter" &:: "operator*" <>$ capt_arg_payload
$--> GenericArrayBackedCollectionIterator.operator_star ~desc:"iterator operator*"
; -"std" &:: "__wrap_iter" &:: "operator++" <>$ capt_arg_payload
$--> GenericArrayBackedCollectionIterator.operator_plus_plus ~desc:"iterator operator++"
; -"__gnu_cxx" &:: "__normal_iterator" &:: "__normal_iterator" <>$ capt_arg_payload
$+ capt_arg_payload
$+...$--> GenericArrayBackedCollectionIterator.constructor ~desc:"iterator constructor"
; -"__gnu_cxx" &:: "__normal_iterator" &:: "operator*" <>$ capt_arg_payload
$--> GenericArrayBackedCollectionIterator.operator_star ~desc:"iterator operator*"
; -"__gnu_cxx" &:: "__normal_iterator" &:: "operator++" <>$ capt_arg_payload
$--> GenericArrayBackedCollectionIterator.operator_plus_plus ~desc:"iterator operator++"
; -"std" &:: "vector" &:: "assign" <>$ capt_arg_payload
$+...$--> StdVector.invalidate_references Assign
; -"std" &:: "vector" &:: "at" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdVector.at ~desc:"std::vector::at()"
; -"std" &:: "vector" &:: "begin" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdVector.vector_begin
; -"std" &:: "vector" &:: "clear" <>$ capt_arg_payload
$--> StdVector.invalidate_references Clear
; -"std" &:: "vector" &:: "emplace" $ capt_arg_payload
$+...$--> StdVector.invalidate_references Emplace
; -"std" &:: "vector" &:: "emplace_back" $ capt_arg_payload
$+...$--> StdVector.invalidate_references EmplaceBack
; -"std" &:: "vector" &:: "insert" <>$ capt_arg_payload
$+...$--> StdVector.invalidate_references Insert
; -"std" &:: "vector" &:: "operator[]" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdVector.at ~desc:"std::vector::at()"
; -"std" &:: "vector" &:: "shrink_to_fit" <>$ capt_arg_payload
$--> StdVector.invalidate_references ShrinkToFit
; -"std" &:: "vector" &:: "push_back" <>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_collection
&::+ (fun _ str -> StringSet.mem str pushback_modeled)
<>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_queue
&::+ (fun _ str -> StringSet.mem str pushback_modeled)
<>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_lang "StringBuilder"
&::+ (fun _ str -> StringSet.mem str pushback_modeled)
<>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_lang "StringBuilder"
&:: "setLength" <>$ capt_arg_payload
$+...$--> StdVector.invalidate_references ShrinkToFit
; +PatternMatch.implements_lang "String"
&::+ (fun _ str -> StringSet.mem str pushback_modeled)
<>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_iterator &:: "remove" <>$ capt_arg_payload
$+...$--> StdVector.push_back
; +PatternMatch.implements_map &:: "put" <>$ capt_arg_payload $+...$--> StdVector.push_back
; +PatternMatch.implements_map &:: "putAll" <>$ capt_arg_payload $+...$--> StdVector.push_back
; -"std" &:: "vector" &:: "reserve" <>$ capt_arg_payload $+...$--> StdVector.reserve
; +PatternMatch.implements_collection
&:: "get" <>$ capt_arg_payload $+ capt_arg_payload
$--> StdVector.at ~desc:"Collection.get()"
; +PatternMatch.implements_list &:: "set" <>$ capt_arg_payload $+ capt_arg_payload
$+ capt_arg_payload $--> JavaCollection.set
; +PatternMatch.implements_iterator &:: "hasNext"
&--> Misc.nondet ~fn_name:"Iterator.hasNext()"
; +PatternMatch.implements_enumeration
&:: "hasMoreElements"
&--> Misc.nondet ~fn_name:"Enumeration.hasMoreElements()"
; +PatternMatch.implements_lang "Object"
&:: "equals"
&--> Misc.nondet ~fn_name:"Object.equals"
; +PatternMatch.implements_lang "Iterable"
&:: "iterator" <>$ capt_arg_payload $+...$--> Misc.id_first_arg
; ( +PatternMatch.implements_iterator &:: "next" <>$ capt_arg_payload
$!--> fun x -> StdVector.at ~desc:"Iterator.next" x (AbstractValue.mk_fresh (), []) )
; ( +PatternMatch.implements_enumeration
&:: "nextElement" <>$ capt_arg_payload
$!--> fun x -> StdVector.at ~desc:"Enumeration.nextElement" x (AbstractValue.mk_fresh (), [])
)
; +PatternMatch.ObjectiveC.is_core_graphics_create_or_copy &++> C.malloc
; +PatternMatch.ObjectiveC.is_core_graphics_release <>$ capt_arg_payload $--> C.free
; -"CFRelease" <>$ capt_arg_payload $--> C.free
; -"CFAutorelease" <>$ capt_arg_payload $--> C.free
; -"CFBridgingRelease" <>$ capt_arg_payload $--> C.cf_bridging_release
; +PatternMatch.ObjectiveC.is_modelled_as_alloc &++> C.malloc_not_null
; +PatternMatch.ObjectiveC.is_modelled_as_free <>$ capt_arg_payload $--> C.free ]
end
let dispatch tenv proc_name args = ProcNameDispatcher.dispatch tenv proc_name args